A Deep Dive into the AI Breakthroughs of 2025

By Graham Sills, Co-Founder & Director of AI

Over the past few years, we’ve witnessed AI evolve from a useful tool to something truly transformational. Each breakthrough has expanded the boundaries of what’s possible. 2025 has been no exception, showing an astonishing rate of change with some really remarkable advancements.

Here at Luminance, we leverage these advances in order to continuously develop our Legal-Grade™ AI to be more powerful than ever before, as we redefine how businesses do work.

For such a momentous year, it’s hard to boil down all the breakthroughs into a few words. But if I had to pinpoint the key developments that are shaping the future of AI – and in turn, Luminance – they would be Context, Reasoning, and Agents. Here’s why they matter…

Context: From Individual Clauses to the Whole Picture

Context windows have continued to grow rapidly, i.e. how much information a model can store in its “short term memory” and process at once. A year ago, a million tokens seemed cutting-edge. Today, some off-the-shelf models can handle two million routinely, with prototypes hitting the 100 million mark. That’s the equivalent of an AI reading and understanding 100 copies of the Bible before answering your question.

At Luminance, we’ve leveraged this progress to move from analyzing clauses in isolation to understanding entire contracts holistically. The result? Smarter insights, more accurate decisions, and a system that truly comprehends your documents in full.

But that’s just the beginning. If you imagine Luminance as the Legal Brain of your organization, then context includes far more than the text within your contracts. Negotiation history, client relationships, budgets, targets, and institutional knowledge – these all impact the way you make a decision. We are using context engineering to bring more and more of these dimensions into the picture. Our north star is a platform that truly understands you: your business, your processes, your strategy. One that anticipates risks, uncovers opportunities, and provides advice grounded in the full picture.

Reasoning: The Why Behind the Answer

Reasoning models mark another huge leap forward. Unlike earlier generations that produced answers without clear logic, these models can now connect the dots, explain their thinking, and justify their conclusions. They unlock the ability to tell the user what is missing when there’s not enough information to give an accurate answer. Reasoning models also allow more complex networks of Agents, but more about that in a minute.

In a legal context, Reasoning models are game-changing. Legal work depends on bringing together multiple pieces of information to make a decision – every conclusion rests on analysis and evidence. A good associate doesn’t just say “yes, they’re liable”; they provide a detailed breakdown of the risks and rationale. Reasoning models unlock swathes of new possibilities for Luminance. Our AI doesn’t just give an answer; it shows its working. This transparency builds trust and makes AI outputs more defensible, auditable, and aligned with how lawyers actually think and act.

Agents: From Isolated Tasks to Autonomous Action 

If 2023 was the year of generative AI, 2025 is the year of AI agents. Last year, agents were simple systems trained for specific tasks, often coordinated by a central orchestration layer. Luminance has used these kinds of agentic systems for a long time, for example when users ask a question across their entire contract repository.

But now, multiple agents can collaborate to complete multi-step tasks with minimal human input. At Luminance, we’re applying this agentic power to the legal world. Our systems can now automatically build workflows, surface documents for approval, generate in-depth research reports, and carry out complex reviews. It’s compound automation with judgment built in.

Of course, as agents become more capable, questions of accountability and control grow in importance. How much autonomy should we give them? Where should humans stay in the loop? We’ve shown how we can automate entire negotiations, but lawyers are not quite ready to take that leap yet. The balance lies in transparency… AI that explains its reasoning and seeks approval at the right moments. And then, as trust grows over time, processes can become fully automated. That’s the reality we’re heading towards – and faster than you might imagine…

If there’s one certainty in AI, it’s acceleration. Every leap opens new possibilities. At Luminance, we’re committed to staying at the forefront of these developments, combining cutting-edge AI with the deep legal expertise our customers rely on. The pace ahead will no doubt be faster than ever, and we couldn’t be in a better position to embrace it.